from version 1.3 beta to version 1.4:
       choose CeCILL-A licence
       adapt PlotSparse for newgraphics
       adapt the interface with minor changes in scilab (lstk->Lstk)
       remove obsolete umfpack_cmplx function.

from version 1.3 alpha to 1.3 beta:
       From a suggestion of Eduardo Lenz
       the functions taucs_chfact and taucs_chsolve work
       also if one provides only the upper triangle of 
       the matrix.

from version 1.2 alpha to 1.3 alpha:

       This new version was motivated by several things:
          * Antonio Manoel Ferreria Frasson sent me some month
            ago a first version adapted for complex matrices
            (concerning the umfpack interface).
          * I am also working currently with complex systems
            (from Helmholtz equation).
          * The likely incorporation of umfpack (with this interface)
            in the future scilab version (3.1 ?); a very good news !
          * update the interface to support umfpack v4.3.

       So many changes occured but most are not visibles from
       the scilab functions provided by this toolbox. For the user
       the main changes are that: x=umfpack(A,"\",b), LUp=umf_lufact(A);
       x = umf_lusolve(LUp,b) can be used with A or b real or
       complex.

        (i)   adapt the Antonio 's modifs (to treat complex linear
              systems) => umfpack_cmplx is obsolete. I have 
              added a field "it" in the cell type of the ListNumeric 
              linked list. It gives the type (real it=0 or complex it=1)
              of the factorisation: this was necessary to choose which
              routines umfpack calling (the xxx_di_xxxxx family (real)
              or the xxxxx_zi_xxxxx family (complex)). Also
              rectangular matrices may be factorised.
    
        (ii)  finally I take the decision to
              change the format (scilab sparse -> CCS) before 
              calling the umfpack routines: this is clearer than 
              the tricks I have done before (one got a factorisation of A' 
              while the interface provides now a factorisation of A).
              This extra work is compensed by the fact that the
              interface is now passed by reference.

        (iii) many changes in the interface C file:
               - I use more heavily the scilab stack in place 
                 of malloc (this simplifies a little all the error 
                 treatment). 
               - Also Francois (Delebecque) tells me about sci_gateway 
                 which avoid the use of PutLhsVar(). 

        (iv)  remove some bugs, in particular in umf_lusolve without 
              refinement.

        (v)   taucs_chsolve with an optional arg (A) do one
              refinement step (=> rafiter is obsolete).

        (vi)  better builder.sce and loader.sce (thanks to Antonio
              Frasson) so that the toolbox is "loadable" from any
              directory.


from version 1.1 alpha to 1.2 alpha:
        (i)   adapt the stuff to umfpack v4.0 

        (ii)  correct a bug in  ReadHBSparse

        (iii) help pages are provided also in html

        (iv)  install explanations may be a little
              clearer than before !   

from version 1.0 alpha to 1.1 alpha :
	(i)   add the taucs snmf stuff interface

	(ii)  add a trick in ReadHBSparse so as to
              read some HB matrices which doesn't respect
              exactly the Harwell-Boeing format !

        (iii) add the 2 macros cond2sp and rafiter and a
              new func res_with_prec

        (iv)  add a trick in PlotSparse to minimize memory
              usage

        (v)   add some tricks in the interface to avoid
              memory leak when CreateVarFromPtr fails
              (see test_size_for_sparse and test_size_for_mat
              in scispt.c).

        (vi)  correct bad path in the .cat files

        (vii) correct a scilab bug (you have to download the
              latest cvs version (begin of December) or to correct 
              yourself, see NOTES 2 in the README file). You must 
              correct it if you want to use umf_luget or taucs_chget 
              with to small sparse matrices.


